Open Garden's first piece in this series laid out how AdCP and ARTF differ. This second installment of our explainer series on agentic AI takes a closer look at AdCP.
AdCP addresses several of the industry's most pressing problems:
Using AI to automate the time-consuming, repetitive tasks of digital marketing
How hard it is to connect platforms and get them talking to each other (through APIs, for instance) to automate those tasks
How complex it is to build AI into the platforms
For those in the back who missed the first episode: AdCP sets a standardized technical spec for building AI agents and making them interoperable across ad platforms. It gives agents a common foundation, so they all work and talk the same way.
Adopting the standard makes digital advertising work easier to automate and cuts the cost of building agents. Companies plug straight into the agent orchestration layer, so deploying an agent gets easier.
AdCP is built as a stack of layers. Each layer sets out roles and a level of tasks.
[1] Business layer
This layer defines roles by use case, mirroring today's ecosystem. Those roles cover the demand side (advertiser, agency, DSP) and the supply side (publisher, SSP).
[2] Orchestration layer
This is the core of the protocol, where AI agents do their work. Each company uses a type of agent tied to the business role it has been assigned. Here's an example.
A sales agent exposes the key details of the inventory a publisher has for sale. For a buyer, it looks up the key details of the inventory to buy.
The agents in the orchestration layer cover the usual ground of online advertising: media planning, audience planning, campaign setup, creative management and more.
[3] Technical execution layer
This layer makes the calls on things such as brand safety and frequency capping, based on platform statistics.
AdCP is designed to run the tasks needed for both direct and programmatic buying.
AdCP in practice
AdCP moves ad buying and selling away from user interfaces and complex APIs toward conversational briefs, for example: "Target sports fans with high purchase intent."
The protocol favors agent-to-agent interactions (A2A / MCP). That opens the door to heavy automation between the buy side and the sell side.
For buyers (agencies, advertisers): automation
Agents can take over the routine work of media trading across the board, for a significant productivity gain:
Media planning
Audience planning
Negotiating volumes, quantities and priorities for deal IDs and direct campaigns
Campaign setup (budget, targeting, line items, creative and so on)
For sellers (publishers): automation
Routine ad ops work gets automated across the board too, for a significant productivity gain:
Negotiating volumes, quantities and priorities for deal IDs and direct campaigns
Campaign setup in the ad server (tags, targeting and so on)
For ad tech platforms (DSPs, SSPs): integration prerequisites
AdCP only works if the buy side and the sell side adopt it at the same time. That's the big challenge. The protocol runs on agents talking to each other and on two-way data exchange, so it succeeds only if it's deployed across the whole operational chain.
Either way, ad tech platforms will have to meet new technical requirements first:
Support the MCP and A2A standards so outside agents can interact with their systems
Build agents natively into their platforms and make them available to buyers and sellers
Assumptions and forecasts
The project is new. The forecasts below are assumptions and will likely be revisited.
Lower costs to build agents
With a technical standard in place, nobody has to build agent systems from scratch. That cuts R&D costs and speeds up the operational rollout of agentic AI.
Efficiency and automation
The protocol replaces the human-to-UI workflow with automation, and agents become the main operators. A 30-minute manual task turns into a single command the agent executes.
Richer signals
AdCP supports flexible inventory and audience discovery through unlimited custom searches. It isn't bound by the traditional OpenRTB taxonomy.
Better transparency
The protocol gives full access to inventory by lifting the limits of bid throttling and sampling. Bid listening data gets easier to analyze, too. Agents aggregate and structure the data points so they're ready to use right away.
The ad tech "tax" and hidden costs
AdCP encourages direct connections between buyers and sellers, which means fewer intermediaries. A cleaner channel cuts hidden costs and the leakage of ad spend along the way.
The rise of "wrappers"
AdCP's open-source foundation will give rise to wrappers, as happened with Prebid. Specialized vendors will build their own tools to customize and extend what the agents can do out of the box.
One down-to-earth reality remains. AdCP won't win because the concept is elegant. It will win if it takes hold in day-to-day operations – if buyers and sellers can actually plug in their agents, exchange clean signals, run complete setups and measure what comes out.
